Fundamentals

AI agent, automation or chatbot: which one does your business need?

Agents, automations and chatbots solve different problems. A plain comparison with examples, a decision checklist and why most setups combine all three.

"We need AI" usually means one of three very different things: an automation, a chatbot, or an AI agent. They solve different problems, cost different amounts, and fail in different ways. Picking the wrong one is the most common reason a first AI project disappoints.

Here is how I explain the difference to clients, and how to choose.

The short version

  • Automation follows fixed rules. When X happens, do Y. It is fast, cheap and predictable, but it breaks when the input is messy.
  • A chatbot or assistant answers questions. It talks to people, usually from a set of documents or FAQs, but it does not change anything in your systems.
  • An AI agent gets work done. It reads messy input, decides what to do, uses your tools, and either finishes the task or hands it to a person for approval.

Side by side

AutomationChatbot / assistantAI agent
Best forFixed, repeatable stepsAnswering questionsTasks that need reading and judgment
InputStructured (forms, fields, events)Questions from peopleMessy (emails, PDFs, messages)
How it decidesRules you writeLanguage model + your contentLanguage model + rules + your tools
Changes your systems?Yes, exactly as toldUsually notYes, with checks and approvals
Typical exampleNew form entry goes to the CRM"What are your opening hours?"Read a supplier email, find the order, update the price
Cost to build and runLowLow to mediumMedium
Main riskBreaks on unexpected inputGives a wrong or made-up answerTakes a wrong action

When automation is enough

If you can describe the task as a flowchart with no "it depends" boxes, you probably don't need AI at all. Tools like Zapier, Make or n8n handle it well:

  • A new booking in your agenda creates a contact in your CRM.
  • An invoice marked as paid sends a thank-you email.
  • Every Monday, a report is pulled from three tools into one spreadsheet.

It is cheap, reliable and easy to maintain. Start here whenever you can.

When you want a chatbot or assistant

If people keep asking your team the same questions, an assistant can answer them from your own content: website, FAQs, procedures, product sheets.

  • Customers asking about prices, opening hours or how a service works.
  • New employees asking how an internal procedure works.
  • Sales asking which product fits which client situation.

The key is grounding: the assistant should answer from your content, show where the answer comes from, and say "I don't know, let me pass you to a colleague" when it is not sure.

When you need an AI agent

Agents are worth it when the input is messy and someone currently has to read it and decide what to do:

  • Email triage. Read every incoming email, understand the request, draft the reply, and route it to the right person.
  • Document processing. Read invoices, quotes or forms in any layout, extract the right fields, check them against your data, and enter them in your system.
  • Follow-ups. Look at open quotes, decide who needs a follow-up and write a personal message.

This is the kind of AI I work on as Head of AI at Herald, in commercial insurance, from reading documents to handling submissions. The same pattern works for a small business, just at a smaller scale and budget.

A simple checklist to decide

Answer these four questions about the task:

  1. Are the steps always the same? If yes, start with automation.
  2. Is the input free text, documents or messages? If yes, you need AI to read it.
  3. Does someone need an answer, or does something need to get done? Answers point to an assistant. Work done points to an agent.
  4. How costly is a mistake? The higher the cost, the more you want a person approving the agent's work, at least at first.

Why the best setups combine all three

In practice, almost every useful system mixes them. Take a wholesaler receiving orders by email:

  • Automation picks up every new email and logs it.
  • An agent reads the email and the attached PDF, understands the order ("same as last month, but 20 boxes instead of 12"), checks stock and prices, and prepares the order in the ERP.
  • The assistant writes a confirmation in the customer's language.
  • A person approves the order with one click, until the system has proven reliable enough to handle routine orders on its own.

Each part does what it is good at. The AI only handles the part that actually needs reading and judgment.

Where to go from here

If you have a task in mind, run it through the checklist above. If you land on "agent", read how to implement AI in your business for the step-by-step approach to a first pilot.

In most cases the honest answer is a mix, and often less AI than people expect.

About the author

Martin Michaux is Head of AI at Herald and was named to Forbes 30 Under 30. Before Herald he co-founded NOX Energy. He writes about what he learns building AI agents in production.

Profile AI workflows More articles
Reply

Book a call with Martin